Computing Norm of Embedding

In your paper, you mention that the L1 norm of the task embedding should correlate with the task complexity.

Do you have sample code of how to compute the norm of the embedding? I can't find it in the repo.

I tried the following approach but can't reproduce the results that were reported.

```python
probe_network = get_model('resnet18', pretrained=True, num_classes=no_classes).cuda()
task2vec = Task2Vec(probe_network, max_samples=5_000, skip_layers=6)
t2v_embed = task2vec.embed(dset)
l1_norm = np.linalg.norm(t2v_embed.hessian, ord=1)
```

Thank you very much!